Are Lakisha and Jamal
More Employable Than
Aisha and Hakim?
A Closer Look at Bertrand and Mullainathan’s
2004 Study “Are Emily and Greg More
Employable Than Lakisha and Jamal?: A Field
Experiment”.
Name: Tom Rosen Jacobson Date: 6/26/2018 Student Number: 11063211
Supervisor: Erik Plug Field: Microeconomics
1 Contents
1. Introduction... 2
2. Literature Overview ... 5
A. 9/11 and Discrimination Against Arabs In The US ... 5
B. Correspondence Tests on Labor Market Discrimination Against Arabs... 6
3. Methodology ... 7
A. Ethnicity Re-Categorization ... 7
B. Testing for differences in Callback Rates ... 11
4. Results ... 12
5. Discussion ... 17
6. Conclusion ... 17
Statement of Originality
This document is written by Thomas Rosen Jacobson, who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
2
1. Introduction
In 2004, Bertrand and Mullainathan’s often cited paper “Are Emily and Greg more employable than Lakisha and Jamal?: A Field Experiment on Labor Market Discrimination” was published in The American Economic Review. The researchers used the correspondence testing method to test for racial discrimination in the labor market in the United States (US), and found substantial differences in callback rates; their employment proxy variable. Though Bertrand and Mullainathan largely succeeded in identifying a causal relationship between callback rate and race, a close inspection of the names chosen for African-Americans to be used throughout the field experiment reveals a potential problem with the studies’ internal validity: some of the names used under the umbrella term “African-American” may sound Arabic1 to some, rather than African-American. Of course, the initial perceived sound of any name is subjective and culturally guided.
Bertrand and Mullainathan did not arbitrarily choose the names used on the fake resumes. Rather, they used a birth certificate database and a survey in subsequent order in an attempt to ensure the names used are representative of the White and African-American ethnicities. Yet despite their methodological control for ethnic representation, first names such as Aisha, Hakim and Rasheed nonetheless sound as if they could belong to Americans of Arabic descent. Furthermore, there is no acknowledgement by the researchers that some names may indeed not sound distinctly African-American.
The timing of the field experiment is also important: though the paper itself was published in 2004, the actual field experiment took place between July 2001 and January 2002 in Boston
1 The minority group “Arabic” is defined here as people that originate from Arabic or middle‐eastern descent.
Furthermore, in line with previous literature such as Arai, Bursell and Nekby (2016), there is no
distinguishment between the terms “Arab”, “Muslim” or “middle‐eastern”. Throughout this thesis, the names thought to sound Arabic will encompass the latter minority groups.
3
and between July 2001 and May 2002 in Chicago, which is incidentally in roughly the same time span as the 9/11 terrorist attacks, and perhaps more importantly, their aftermath.
Therefore, it is clear that there is a need to examine whether or not a substantial portion within the callback rate differential across the main findings is attributable to the idea that some of the African-American names used are actually perceived as Arabic. If employers indeed perceive some of the names as Arabic, it could be that employers are, at times, discriminating against Arabs2, rather than against African-Americans. This tilted discrimination is expectable considering the study’s coincidence with the 9/11 attacks. Therefore, this thesis will attempt to answer to what extent discrimination against African-Americans as found in Bertrand and Mullainathan’s 2004 study “Are Emily and Greg More Employable than Lakisha and Jamal?: A Field Experiment” is attributable to discrimination against Arabs.
The importance of examining this potential validity issue is twofold: first, if a portion of the variation in callback rates between Whites and African-Americans is attributable to variation in callback rates between African-American names that ought to be classified as Arabic, then researchers must be made aware of this currently overlooked internal validity issue. Second, if a causal interpretation can be given to any potential callback rate differentials regarding Arabic names, this implies the methodology used to determine Arabic names could be of interest to researchers looking for new ways to find evidence of discrimination in the labor market.
Ideally, Bertrand and Mullainathan’s method of identifying and verifying first names to be used in their study would be replicated perfectly while incorporating Arabic names into
2 As Arai and Skogman Thoursie (2009) and Arai, Bursell and Nekby (2016) show, there is evidence of
4
the dataset. However, the original method is undeniably time consuming and requires resources, to some extent, to imitate properly and see if there are indeed possible names that can be classified as “Muslim”, rather than “African-American”. Thus, a need arises for a different methodology that may be followed given relatively few resources and time constraints. Therefore, using the online Muslim community to see whether or not names on the list of Bertrand and Mullainathan are indeed considered Muslim is a viable alternative.
The extent to which the timing of the study did indeed influence the results, will determine how severe this potential problem of internal validity truly is for Bertrand and Mullainathan’s work. Moreover, in estimating employment differentials, the study would need to acknowledge that they use a smaller set of observations per ethnic minority than previously stated. This is important as a change in the distribution of the number of observations across race could have implications for the significance of results in the original study. In the extreme case that the estimates of employment differentials, given the new selection of names and number of observations per race are no longer significant, as a result the original study would be unsuccessful in identifying differential treatment by race in the labor market. Considering the effect 9/11 had on public perception of stereotypes of Muslims as religious extremists it is hypothesized that if there are any African-American first names, that are actually Arabic-sounding, these will, on average, receive equal or fewer calls back on job applications than the African-American categorical average.
In order for results between the original study and this thesis to be comparable, the statistical methodology used by Bertrand and Mullainathan (2004) for their findings will be replicated. The goal is to examine whether there is a need to address internal validity issues in the original study. Additionally, the methodology used is evaluated on its ability to identify differential treatment in the labor market, using the dataset of the original study. In order to
5
do this: first, in section 2, Bertrand and Mullainathan’s study in reference to the 9/11 attacks is contextualized. Second, in section 3, an explanation is provided for the methods and statistical tests used to identify Muslim-sounding names amongst the African-American names initially used and test for Muslim-African-American employment differentials. Third, in section 4, the results of the current paper are provided. Fourth, in section 5, results are evaluated and contextualized and fifth, in section 6, concluding remarks are provided.
2. Literature Overview
A. 9/11 and Discrimination Against Arabs In The US
According to The Arab American Institute Foundation (2014), there were around 2.6 million people in the US that identified as having Arab-speaking ancestry in the year 2000. Furthermore, as of 2012, 94% of the Arab population live in high density areas, including Chicago and Boston. The Arab population in 2000 thus represented just below 1% of the total US population3, and just below 4% of the total minority population4.
Disha, Cavendish and King (2011) write extensively about the pattern of anti-minority hate crimes, both prior to and after 9/11. For instance, they find that in the 10 years prior to 9/11, anti-Arab hate crime remains relatively stable. Furthermore, the anti-Arab hate crime trend line follows a pattern similar to “Anti-black”, “Anti-Other Ethnicity” and “Anti-Jewish” trend lines. However, shortly after 9/11, an immediate surge in anti-Arab hate crimes is documented. This is in line with a report released by The Federal Bureau of Investigation (FBI) (2002), which stated an increase in anti-Arab hate crimes between 2000 and 2001 of
3 Source: World Bank (2018) 4 Source: 2000 United States Census
6
1600%5. The researchers examine in detail how 9/11 has affected anti-Arab hate crimes. They find that, both before and after 9/11, most hate crimes against Arabs take place in areas where the fraction of Arabs in the area is largest.
Barbara Perry (2001, 2002) and Liu-In Wang (2002) find that shortly after the events of 9/11 the extent to which anti-Arab discrimination takes place was intensified and demonstrate how anti-Arab rhetoric and anti-Arab hate crimes have become “normalized”. They argue that the hostile climate created against Arabs shortly after the 9/11 attacks lead to the American ethnic majority feeling increasingly united against a “new enemy”.
B. Correspondence Tests on Labor Market Discrimination
Against Arabs
Some labor market discrimination studies using the correspondence testing method have tested for discrimination against Arabs. For instance, Arai and Skogman Thoursie (2009) test for discrimination against Arabs in Sweden at the extensive margin. They exploit variations in wage that arise from people originating from ethnic minorities who have changed their surnames throughout the 90’s to a Swedish sounding one. They find that there is, on average, a significant increase in wages after the name change has taken place for people of african and asian descent. They argue that this increase in wage is attributable to discrimination against people of african and or asian descent, as perceived by employers. However, they acknowledge that there is no empirical evidence to support the claim that name changes do not coincide with omitted variables such as higher productivity or more intense job searching, occurring at the time of the name change. Moa Bursell argues that people of middle-eastern descent living in Sweden that change their surname do so in order to circumvent the minority penalty that must be paid when entering the labor market (2012).
7
Arai, Bursell and Nekby (2015) study the reverse gender wage gap among ethnic minorities in Sweden, using a resume correspondence testing method. They find evidence of a reverse gender wage gap, and state that their results are in line with the Outgroup Male Target Hypothesis, as initially presented by Navarette et al. (2010). Interestingly, this would imply that, contrary to popular belief, the idea of a “double burden” on minority females is not empirically supported (Arai, Bursell and Nekby, 2015). The study concludes by stating that minority groups are discriminated against through 2 avenues: discrimination in the labor market and differences in minority stereotypes across genders.
3. Methodology
A. Ethnicity Re-Categorization
This thesis uses only the database provided by Bertrand and Mullainathan. In order to be sure that there are no inaccuracies when re-purposing the data, it is important to guarantee that the relevant aspects of Bertrand and Mullainathan’s study are reproducible. This comes down to reproducing the key tables of the study. Table 1 in the appendix presents the attempt at reproducing the table 1 of the original study. As seen in the table, not all the results are identical. It is important to note that Bertrand and Mullainathan’s data files do not include any step by step instructions on the reproduction of their results. Resultantly, it is difficult to understand the exact interpretation of joint variables. For example, though there are many proxy variables in the dataset that represent specific occupations and industries, there are no interaction proxy variables labelled “Sales jobs” or “Administrative Jobs”. After attempting to reproduce rows five and six using trial and error, it was decided that these results are not strictly relevant for the purposes of this thesis. As such, reproducing this part of the study has not taken place.
8
After confirming the main results, it is important to create a method that structurally determines which African-American names can be considered Muslim, and which to exclude. There are a number of different methods available for exploration. The first and most obvious approach would be to mirror Bertrand and Mullainathan’s method of determining which of the initially drafted names were considered African-American and which were considered White. However, due to the purpose of this thesis, this method inevitably leads to complications: Bertrand and Mullainathan select names based on name frequency data which is taken from birth certificates, created between 1974 and 1979 in Massachusetts. They then tabulate the data and select names that are used mostly for African-American and White newborns. In order to mirror this process as accurately as possible, the original Massachusetts birth certificate database would first need to be re-examined to check whether there are any mention of minorities such as “Muslim” within the umbrella term “African-American” or if there is a distinction between these two labels. If this is the case, one may argue that Bertrand and Mullainathan’s definition of “African-American” is inaccurate. This may result in part of the variation in callback rates being attributable to employers actually discriminating against other minorities, and not solely against African-Americans.
Assuming the database distinguishes between many different minorities, the next step is to examine how Bertrand and Mullainathan checked if the names on the shortlist indeed are relevant at the time of their study. To check if these names are indeed perceived as African-American, the researchers use a survey. They receive 30 responses per name, whereby the respondents answered what race each name they believe belongs to7, among other features.
One may argue that 30 responses per name seems insufficient. This is because the study takes place in two densely populated areas: Chicago and Boston. It should also be noted that the
9
survey was conducted in Chicago. Thus, if Boston locals do not perceive first and last names in the same way as Chicago locals do, this may influence the final list of names. In short, using only 30 responses per name taken from a survey conducted in only 1 of 2 locations may not be enough to confirm that the current names on the list sound the most distinctly “White” or “African-American” out of all potential options.
Whether or not the methodology of Bertrand and Mullainathan has weaknesses, it is undeniably time consuming and requires resources, to some extent, to completely reproduce properly and see if there are indeed possible names that can be classified as “Muslim”, rather than “African-American”. Thus, a need arises for a different methodology that may be followed given relatively few resources and time constraints. Therefore, using the online Muslim community to see whether or not names on the list of Bertrand and Mullainathan are indeed considered Muslim seemed to be a viable alternative.
There are many different online forums where Muslim parents post threads relating to life as a Muslim in western society, specifically Europe and the United States. As such, there are many threads on these forums in which users ask the general question “What are some Muslim names that sound English?”. As a consequence of the large amount of responses to threads relating to this question, the opportunity arises to cross reference the list of “African-American” names of both genders with the number of mentions those names receive on the forums. However, before being able to rank any names at all, a method needs to be devised to structurally group African-American names into 2 different categories: “Muslim” and “distinctly African-American”. In order to do this, a website was called upon that contains a thorough list of many commonly used Muslim names. Using the search tool in the website, names found on the list of Bertrand and Mullainathan’s table 8 were searched for. Of the 18 names entered, 5 received a hit on the website. The other 13 names were excluded from the
10
analyses of this thesis. Throughout the search process it also became clear that there are many different variations of Muslim names that differ in spelling but are pronounced exactly the same. If one assumes that a name connected to Islam is closely related to its pronunciation, it is reasonable to argue that spelling variations on otherwise similar names can be seen as virtually identical to the base name. In this thesis the latter is assumed. That is, the hits that the variations received are included in the total number of hits of their respective base names.
After identifying which names sound Muslim and which do not, it is also necessary to see how Muslim each of the names sound as perceived by the Muslim community. This is important, because although there is now a theoretical list of Muslim-sounding names, it must be confirmed that these names are indeed perceived as Muslim sounding, preferably provided by people living in a western society. Furthermore, by quantifying this measure, one could attempt to estimate discrimination against Arabs at the intensive margin, as well as the extensive margin.
15 different threads found on forums relating to Muslim parental life in the west were selected. Within these threads the online search function (ctrl + f in the browser) was used to search for the five selected names on the list. After the number of hits across these websites were tallied, they were ranked from least to most hits. This is because the threads specifically provided Muslim names that sound English. Thus, by ranking them from least to most, the least English sounding Muslim names used in the west (i.e. the most Muslim sounding names used in the west) are ranked highest. As an extra check, names that got 0 hits would be excluded from further analysis. However, all 5 names received more than 0 hits. Table B2 in the appendix presents the results.
After the final list of names was created, a series of statistical tests were run on them. These include simple OLS regressions and tests for significance for the differences between
11
the average callback rates of distinctly Muslim names and distinctly African-American names. Essentially, if there is a statistically significant difference between these values that persists throughout multiple names, this could be explained by the Islamic relatedness of the names.
B. Testing for differences in Callback Rates
First, the variable “race” now holds an additional category, namely the Arabic minority. Because we want to test whether the difference in callback rates between Arabs and African-Americans is apparent at both the extensive and the intensive margin, we want to test specifically for statistical significance of the difference in the callback rates, as well as for the average callback rate of Arabs being higher or lower than the average African-American callback rate. We want to test this using student’s t-tests. In order to accomplish this, we must first take a step back and see if the variances in the callback rates between African-Americans and Arabs are (un)equal. An F-test for equal variances rejects the null hypothesis (p-value=0.0000). Similar to Bertrand and Mullainathan (date), we assume sample independence at the first name level. Thus, when testing for statistical significance in outcome variables, an unpaired two-sample t-test will be used.
Super impressed by your methodology, really cannot say much about it!
Discrimination against Arabs at the intensive level is also tested. We describe discrimination against Arabs at the intensive margin as follows: resumes that yield relatively more Arabic-sounding names will, on average, generally have lower callback rates. Then, using the number of hits received across all the forums found in appendix X, we can rank the names from Arabic names that sound most to least English and use a spearman rank correlation coefficient to estimate the relationship.
12
C. Weaknesses of the Current Methodology
The ability of a methodology to estimate discrimination in the labor market depends on its ability to identify a causal relationship between the applicant’s perceived race and the corresponding callback rate; it must be proven that employers, on average, make fewer calls back to Arabic/African-American sounding names because of their perceived race. On the one hand, the current categorization methodology is successful at identifying first names that are perceived as either distinctly Arabic-sounding or distinctly African-American sounding. On the other, there are many potentially omitted variables that are not addressed. For instance, the last names on the fictitious resumes sent out during the original field experiment are not accounted for, which is likely to be a major source of inaccuracy. This is because, though some first names can be perceived as Arabic sounding, many of the last names used are unambiguously African-American. This is clear when looking at section II.B. in the footnotes of the original study: surnames such as Jackson, Jones and Robinson are almost certainly exclusively African-American. This means that, although this thesis succeeds in finding differential treatment across first names, there is no evidence that this difference persists when controlling for surnames. As found by Arai and Skogman Thoursie (2009), there is an employment penalty attached to having a Asian/African/Slavic surname in Sweden. Though this study was not conducted in the US, these findings still suggest that not controlling for surnames is a likely source of omitted variable bias. Further research on Bertrand and Mullainathan (2004) can be conducted where statistical analysis is done while controlling for the effect of having a distinctly African-American surname. Another limiting factor is that this methodology, unlike the original study, does not account for the name frequency distribution of the Arabic names. Though the current methodology offers an alternative method to estimate name frequencies by using a contemporary platform, this method does not come without inherent drawbacks: first, consulting online platforms
13
Table 1.a. – Mean Callback Rates by Racial Soundingness of Names
Callback rate (%) Differences (%)
White AA Arab Ratio (White/ AA) Ratio (AA/Arab ) Ratio (White/AA) AA-White Arab-AA Sex M 8.87 6.62 4.86 1.27 1.37 1.84 -2.25* -1.76* F 9.89 7.09 2.22 1.39 3.22 4.45 -2.80* -4.87* City Chicago 8.06 5.77 3.14 1.39 1.85 2.57 -2.29* -2.63 Boston 11.63 8.74 4.34 1.33 2.00 2.68 -1.01* -4.50* Overall 9.65 7.02 3.75 1.37 1.87 2.57 -2.63* -3.27*
● *=2-tailed significance at 5% level. ● AA stands for “African-American”.
implies that all names re-categorized as Arabic come from an online source. Because the internet only became accessible for everyday purposes in the mid 90’s, all comments suggesting “Muslim names that sound English” presumably stem from parents whose children were born around or after this time span. If this is true, then the average age of the Arab population around which the current name-distinction method is based is likely below the average age used on the fake resumes in the original studies. As such, the first names identified as Arabic-sounding may be biased towards a much younger age demographic.
Another reason why this method of re-categorization of first names across ethnicities does not give an accurate approximation of the actual Arabic name frequency distribution in the US is because any user across the globe can respond to the online threads, not just Americans, let alone Chicago or Boston locals.
4. Results
Table 1.a. presents the first part of the main findings. Looking at the overall outcomes, it can be seen that the difference in percentage number of calls back between Whites and African-Americans is negative and statistically significant. This suggests that, despite the new distinction between Arabs and African-Americans within the “race”
14
variable, resumes with an African-American first name still receive less callbacks than resumes with a White first name. Specifically, under the new categorization of races, African-Americans receive, on average, 37% fewer calls back than Whites. Interestingly, this callback rate differential is 13 percentage points lower than 50%, as found in the original paper. This raises the question of to what extent distinctly African-American names benefit from the new race-categorization.
Table 1.b. below is extends the statistical analysis of table 1.a. across the industry dummy variables included in the original study. Surprisingly, the only significant callback rate differential between African-Americans and Arabs is found in the Public sector. Here, resumes with an Arabic-sounding name are 415% less likely to receive a call back than similar resumes with American names. The callback differential between African-Americans and Whites under the new first name categorization method differs significantly in the F.I.R.E. (Financial, Insurance and Real Estate ) sector and in the Business and Personal Services sector. In the F.I.R.E. industry, resumes with an Arabic-sounding first name receive 30% less calls back than similar resumes with an African-American-sounding first name. Though the difference is insignificant, it is noteworthy that the only positive callback difference between African-Americans and Whites is found in the Transport industry.
15
Table 1.b. – Mean Callback Rates by Racial Soundingness of Names per Industry
Callback Rate (%) Differences (%)
White AA Arab Ratio
(White /AA) Ratio (AA/Arab ) AA-White Arab-AA Industry Dummy Manufa cturing 6.93 4.35 2.44 1.59 1.78 -2.58 -1.91 Transpo rt 12.17 17.54 5.88 0.69 2.98 5.37 -11.66 F.I.R.E. 10.14 4.49 3.45 2.26 1.30 -5.65* -1.04 Wholesa le/Retail 8.64 5.78 3.25 1.49 1.78 -2.86 -2.53 Busines s 10.43 6.74 3.88 1.55 1.74 -3.69* -2.86 Public 11.41 10.12 2.44 1.13 4.15 -1.29 -7.68* Other/U nknown 8.71 6.69 5.48 1.30 1.22 -2.02 -1.21 Overall 9.65 7.02 3.75 1.37 1.87 -2.63* -3.27*
● *=2-tailed significance at 5% level
● For Arab sample, number of observations per outcome is displayed in parentheses
It is possible to test for correlation between the degree to which an Arabic name commonly given to newborns in the west is Arabic-sounding and their respective callback rates. Using the current methodology in which the gross number of hits on each name identified as Arabic, Spearman’s ranked correlation coefficient can be estimated. The results are shown in Table 2, where names are ranked based on the number of hits each name received across 15 forums. Testing for Spearman’s coefficient may give insight into the ability of the methodology to estimate callback rate differentials in the US labor market against Arabs at the intensive margin.
16
Table 2 – List of Arabic Sounding Names Ranked In Reverse Order of “English-Soundingness”
Rank Name Callback Rate (%)
1 Hakim 5.46
2 Rasheed 2.99
3 Kareem 4.69
4 Aisha 2.22
5 Jamal 6.56
The spearman rank order correlation coefficient is 0.1, meaning the null hypothesis that there is no relation between the inverse English-soundingness of a Muslim name and the callback rate is not rejected.
Because the set of first names that are perceived as African-American differs between this thesis and the original study, it is possible to compare the effect of re-categorizing the set of first names on the callback rate on resumes with African-American names. This is done in table 4. It is found that the change in the set of names labelled as African-American across the 2 studies results in statistically significant differences between the callback rates of African-American sounding names: overall, resumes with distinctly African-American-sounding names (current assignment of race across names), are, on average, 9% more likely to receive a call back that names that are non-distinctly African-American (Bertrand &
17
Table 4– Mean Callback Rate Differentials (%) for African-Americans per Race Categorization Method
Base Male Female Chicago Boston
New (White, A, Arab) 7.02 6.62 7.09 5.77 8.74
Old (White, A) 6.45 5.83 6.63 5.4 7.76
Ratio 1.09* 1.14* 1.07* 1.07* 1.13*
Difference 0.57 0.79 0.46 0.37 0.98
*=2-tailed significance at 5% level
Mullainathan (2004) assignment of names). Males benefit the most from this distinction: resumes with a distinctly African-American male name receive 14% more calls back, while their female counterparts only receive 7% more calls back. Furthermore, resumes that are perceived as distinctly African-American receive 7% more calls back in Chicago while their Boston counterparts receive 13% more calls back. This suggests that Chicago employers are, of the 2 cities, least sensitive to distinctions amongst racial minorities. Notably, though the initial difference between the callback rates on African-American names is on average 13 percentage points, the results in Table 2 show that only an average of 9 percentage points (69%) of the latter 13% difference is attributable to the new method of first name categorization across race.
5. Discussion
A. Identifying Causal Relationships
In line with aforementioned literature that employ the correspondence testing method, causal interpretation is are given to any significant callback rate differentials that involve exclusively between-race and intra-race differences amongst the “White” and “African-American” ethnicities. This is possible because the callback rate differentials reported for the latter 2 groups are found using the exact same data analysis tools as Bertrand and
18
Mullainathan originally used when reporting on the specified outcomes. Hence, if one wishes to critically examine the supporting arguments for causal interpretation of these results, I refer to section IV.A. and IV.B. in Bertrand and Mullainathan (2004). On the other hand, the set of resume first names that sound Arabic are identified using a methodology that differs from the original study. Therefore, any causal interpretations regarding Arabic-sounding names must result directly from the ability of the current methodology to estimate causal relations.
B. Interpretation of Results
Looking at tables 1.a., a clear pattern emerges: the callback rate differential between Arabic-sounding names and African-American-sounding names is structurally below 0. This is suggestive of discrimination to a higher degree against Arabs than discrimination against African-Americans in Chicago and Boston, which is in line with the hypothesis put forward in section 1. A theory for this result that puts discrimination against Arabs in the US into an appropriate context for analysis is as follows: using the average number of anti-Arab hate crimes per capita within a specific area as a proxy for discrimination at the intensive level and assuming employers are generally representative of the average ethnic majority citizen in their area, then, if one considers where the fraction of Arabs in a population is highest (i.e. Chicago and Boston) and the spike in anti-Arab hate crimes shortly after 9/11, it is arguable that employers in places such as Chicago and Boston discriminated particularly strongly against Arabs. Considering that Disha, Cavendish and King (2011) report a major spike of anti-Arab hate crimes that took place right after the field experiment was initiated in both cities, employers may have developed a disproportionately strong distaste for Arabs in the last 3-4 months of 2001 and the first few months of 2002.
19 A complication arises when taking into account that the resumes with Arabic-sounding first names all had distinctly African-American Arabic-sounding surnames. This means that, if employers indeed had an extraordinary distaste against Arabs during the time of the field experiment, this distaste was so strong that employers were triggered into responding in a discriminatory way merely by the possibility that the applicant was Arabic. Another explanation is that employers pay more attention to first names than they do to surnames. However, this may not be in line with Arai and Skogman Thoursie (2009).
Table 1.b. reports mean callback rates per race across industry. Here a potential drawback of increasing the number of subcategories within a categorical value is revealed: a potential non-large sample pool of the newly added subcategory. Essentially, because there are only 5 names out of the total 18 African-American-sounding names as given by the original study, the number of observations for the new subcategory “Arab” is significantly lower than the number of observations for “Distinctly African-American” and “White”. This in and of itself does not necessarily lead to statistical inaccuracies. However, as evidenced by the difference in significant callback rate differentials between Arabic and African-American sounding names, standard errors of the differentials increase substantially when the total number of observations is spread thin across too many different dummy variables. Further research that studies the effect of race on callback rate differentials between the ethnic American majority and Arabs may alleviate this issue.
Table 2 reports callback rates per name, ranked from least English-sounding to most English-sounding. The Spearman correlation coefficient is too small to reject the Spearman null hypothesis. A potential explanation for this is that employers themselves do not discriminate more against that names that are more Arabic-sounding than others. Rather, employers merely discriminate across race in varying degrees, but do not discriminate within a minority against some more than others.
Table 3 reports the effect of the change in set of names that are considered African-American-sounding per study. The mean callback rates across all factors is higher in the
20 current study than in the original. This suggests that making a distinction between African-American-sounding first names and Arabic-African-American-sounding first names is important. In addition, the results are suggestive of the idea that Bertrand and Mullainathan’s set of names used for the African-American ethnicity also encompasses (part of) the Arabic-American ethnic minority. Though causal interpretation cannot be given to any results including the Arabic-American ethnic minority, further research involving a better method of creating distinctions between minorities based on (first) names may find interesting results.
6. Conclusion
This thesis hypothesized that the callback rate differential between Arabic-sounding names and African-American sounding names would be equal to or below 0. It is found that, although the hypothesis has predicted some the main outcomes correctly, the characteristics of the current methodology prohibit giving causal interpretations to any outcomes that involve the set of names defined as Arabic-sounding. It has also not found any relation between how non-English-sounding an Arabic name is and a lower callback rate. On the other hand, the results are suggestive of the need to create a distinction between African-American-sounding names and Arabic-sounding names in the original dataset of Bertrand and Mullainathan (2004). In summation, though the current methodology was somewhat successful in highlighting potential issues with Bertrand and Mullainathan’s approach to conducting a correspondence test, it is unable to support any claims that the original study must be amended with causal interpretations of empirical findings.
21
7. References
Arai, M., Bursell, M., & Nekby, L. (2015). The Reverse Gender Gap in Ethnic Discrimination: Employer Stereotypes of Men and Women with Arabic Names,. International Migration Review, 50(2), 385-412. doi: 10.1111/imre.12170
Arai, M., & Skogman Thoursie, P. (2009). Renouncing Personal Names: An Empirical Examination of Surname Change and Earnings. Journal Of Labor Economics, 27(1), 127-147. doi: 10.1086/593964
Altonji, J., & Blank, R. (1999). Race and gender in the labor market. Handbook Of Labor
Economics, 3, 3143–3259.
Banerjee, A., Bertrand, M., Datta, S., & Mullainathan, S. (2009). Labor market discrimination in Delhi: Evidence from a field experiment. Journal Of Comparative Economics, 37(1), 14-27. doi: 10.1016/j.jce.2008.09.002
Bertrand, M., & Mullainathan, S. (2004). Are Emily and Greg More Employable than Lakisha and Jamal? A Field Experiment on Labor Market Discrimination. SSRN Electronic Journal. doi: 10.2139/ssrn.422902
Bursell, M. (2012). Name change and destigmatization among Middle Eastern immigrants in Sweden. Ethnic And Racial Studies, 35(3), 471-487. doi: 10.1080/01419870.2011.589522
Deming, D., Yuchtman, N., Abulafi, A., Goldin, C., & Katz, L. (2016). The Value of Postsecondary Credentials in the Labor Market: An Experimental Study. American Economic Review, 106(3), 778-806. doi: 10.1257/aer.20141757
Disha, I., Cavendish, J., & King, R. (2011). Historical Events and Spaces of Hate: Hate Crimes against Arabs and Muslims in Post-9/11 America. Social Problems, 58(1), 21-46. doi:
22 Goldin, C., & Rouse, C. (2000). Orchestrating Impartiality: The Impact of “Blind” Auditions on Female Musicians. American Economic Review, 90(4), 715-741. doi: 10.1257/aer.90.4.715
Navarrete, C., McDonald, M., Molina, L., Sidanius, J., & Simpson, J. (2010). Prejudice at the nexus of race and gender: An outgroup male target hypothesis. Journal Of Personality And Social
Psychology, 98(6), 933-945. doi: 10.1037/a0017931
Perry, B. (2001). In the name of hate: understanding hate crimes. Choice Reviews Online, 39(03), 39-1899. doi: 10.5860/choice.39-1899
The Arab American Institute Foundation. (2018). Demographics (p. 1). Washington, DC: The Arab American Institute Foundation. Retrieved from http://www.aaiusa.org/demographics
Wang, L. (2002). Hate crime and everyday discrimination: influences of and on the social context. Rutgers Race And The Law Review, 4(1), 1-31.
23
Appendix 1
Table 1: List of Names and Determination of Arabic-origin Name Observed
(1=Yes)
Variations
Females Aisha 1 Aishah, Ayesha, Aiesha, Aaisha, Aisia, Aisiah, Ayeesha, Ayeesa, Ayeeshah, Ayeisha, Ayisha
Keisha 0 Tamika 0 Lakisha 0 Tanisha 1 Latoya 0 Kenya 0 Latonya 0 Ebony 1 Abnus
Males Rasheed 1 Rashid, Raashid, Rashed
Tremayne 0
Kareem 1 Karim, Kaarim
Darnell 0
Tyrone 0
Hakim 1 Hakeem, Hakem
Jamal 1 Jamaal, Jamail*
Leroy 0
Jermaine 0
24
Appendix 2
Name Male Female Source Rasheed/ Rashid Kareem/ Karim Hak im Jamal/Jamail /Jamaal Aisha/a isyah http://www.brandedgirls.com/english-muslim-names/ https://www.netmums.com/coffeehouse/becoming-mum- pregnancy-996/baby-names-643/1226050-english-sounding-muslim-names.html 1 1 https://www.ummah.com/forum/forum/misc/anonymous- posting-counselling-forum/139606-nice-muslim-names-that-sonund-english 1 1 111 https://www.mumsnet.com/Talk/baby_names/804247-Oh-help-English-sounding-muslim-boys-names?messages=100&pg=1 1 1 1 http://www.emmasdiary.co.uk/forums/baby-names-forum/british-sounding-muslim-names-t214751.html https://community.babycentre.co.uk/post/a27983443/british_sou nding_muslim_names... https://community.babycenter.com/post/a65663896/muslim-names-that-sound-western 1 1 1 https://www.whattoexpect.com/forums/blended-and-multicultural-families/topic/muslim-names.html?page=2 1 1 111 1111 https://nameberry.com/nametalk/threads/153027-Muslim- names-that-are-easy-in-English/page2?s=6b341a5d70854b77d08cde83a558057c NEW SEARCH COMMAND: "Arabic names that sound english"
https://www.mumsnet.com/Talk/baby_names/1465870-Arabic-yet-English-sounding-boys-names 11
https://community.babycenter.com/post/a68287675/arabic-baby-name-that-sound-english. Rashad 1 1, Jamil 1, Jamel 1 https://www.whattoexpect.com/forums/june-2011-babies/topic/arabic-baby-boy-name-that-sounds-american.html 1 http://www.twinstuff.com/forums/threads/looking-for-arabic-names-that-sound-good-in-american-english-too.41415/ http://www.britishbabynames.com/blog/2012/01/arabic-names.html 1 1 1 1, Jamil 1 1 https://community.babycentre.co.uk/post/a26303793/boys_musl imarabic_names_that_sound_english TOTAL (GROSS) 6 6 2 11 10 TOTAL (NET) 5 6 2 8 10
